metadata

task_categories:
  - summarization
language:
  - en

Dataset Card for processed_dataset_top.csv

This dataset is an enhanced version of the CNN/DailyMail summarization dataset. Articles have been preprocessed and keywords are prepended at the top of each article to provide additional context for fine-tuning summarization models.

Dataset Details

Dataset Description

The dataset includes news articles with keywords prepended at the top, formatted with special tokens for compatibility with transformer-based models. Keywords were extracted using KeyBERT to emphasize key topics from the articles. Each article is paired with its corresponding summary (highlights). Dataset Sources Original Dataset

The original dataset is the CNN/DailyMail summarization dataset, which contains:

Articles: News articles from CNN and DailyMail.
Highlights: Human-written summaries of the articles.

Preprocessing Applied

Keyword Extraction:

Extracted keywords using KeyBERT.
Keywords were formatted with <keyword> special tokens and prepended at the top of each article.

Dataset Structure

The dataset contains two columns:

article: Preprocessed articles with keywords prepended at the top.
highlights: Preprocessed summaries (highlights).

Example:

Article:

Keywords: GLOBAL ECONOMY, INFLATION, SUPPLY CHAIN

The global economy is facing unprecedented challenges due to inflation and supply chain disruptions.

Highlights:

Global economy faces challenges from inflation and supply chain issues.

Intended Use

This dataset was created to provide an enhanced summarization dataset for experiments in keyword-based summarization. Prepending keywords at the top of the text acts as a primer, potentially improving model performance by focusing attention on key topics early.

Possible Use Cases:

Fine-tuning summarization models such as DistilBART or BART.
Evaluating the impact of prepended contextual keywords on summarization accuracy.

Limitations

Contextual Bias:
    Prepending keywords may introduce a bias where the model overly focuses on the prepended keywords rather than the article's main content.

Keyword Extraction Quality:
    Automatically extracted keywords might not always reflect the true focus of the article.

Citation

If using this dataset, please cite the original CNN/DailyMail summarization dataset and mention the preprocessing and keyword extraction enhancements.